Inside the $23M AFT-OpenAI-Microsoft AI Deal

In July 2025, four months before ChatGPT for Teachers went live, the American Federation of Teachers stood on a stage with three technology companies and announced something unusual: a jointly funded, $23 million effort to train 400,000 K-12 educators on artificial intelligence over five years. The National Academy for AI Instruction is the scaffolding underneath the free-teacher-tool moment — the workforce program that makes a product launch feel less like a marketing push and more like an industry laying down infrastructure. If you want to understand why OpenAI is giving away a premium product to teachers, the money that funds this academy is where the logic becomes visible.

Deals like this are easy to wave past as a press release with a big number attached. But the structure of the funding — who pays, in what form, over what timeline, aimed at what target — tells you what each party actually expects to get. So let's read the ledger.

The partners at the table

The National Academy for AI Instruction has five named founders, and it's worth separating the roles. On the labor side: the American Federation of Teachers, which represents roughly 1.8 million members, and its New York City local, the United Federation of Teachers. On the technology side: OpenAI, Microsoft, and — a detail that surprises people — Anthropic, the maker of Claude. Three competing AI labs agreed to co-fund the same teacher-training program through the same union.

That last fact reframes the whole thing. This isn't OpenAI buying exclusive access to a union's membership; it's a shared-cost coalition where direct rivals sit on the same funding line. We unpack that strange triangle more fully in three rivals, one classroom, but even at the level of the balance sheet it changes the read: no single vendor owns the curriculum, because no single vendor paid for all of it.

Following the $23 million

The headline figure is $23 million in total funding across the five-year life of the initiative. The piece of it that's been detailed publicly is OpenAI's, and its shape is instructive:

• $10 million total from OpenAI, split into two very different kinds of contribution.
• $8 million in direct funding — actual dollars into the program.
• $2 million in-kind — not cash, but engineering support, compute access, and technical guidance, per OpenAI's own account of the partnership.

The remaining roughly $13 million comes from the other technology partners, Microsoft and Anthropic. (OpenAI's is the split that's been laid out line by line; treat the balance as the combined contribution of the other two labs rather than a precisely published per-company figure.)

That $8M-plus-$2M structure is the part worth pausing on, because in-kind funding is where incentives hide. When a company donates cash, it buys goodwill and a seat at the table. When it donates compute, engineering time, and technical guidance, it also embeds its own tools and its own people in the room where the curriculum gets built. The $2 million in-kind isn't a rounding error on the $8 million — it's arguably the more strategic half, because it shapes what teachers learn to use, not just that they get trained. A teacher trained on OpenAI-supported infrastructure is a teacher fluent in OpenAI's product surface.

The cash builds the academy. The in-kind contribution — compute, engineers, technical guidance — builds the habits. One is a donation; the other is a distribution channel.

None of that makes the in-kind support improper. Training teachers on real, capable tools requires real, capable tools, and someone has to supply the compute. But it's the mechanism by which "we funded teacher training" quietly becomes "we shaped how teachers were trained," and it's why the funding structure deserves more scrutiny than the topline number invites.

Five years, 400,000 teachers, one physical anchor

The scope is defined by three numbers, and they compound. Five years. 400,000 educators — which reporting on the launch frames as roughly one in ten U.S. teachers. And a physical footprint that starts with one building and is meant to grow.

The flagship hub opened in Fall 2025 inside the UFT's downtown Manhattan headquarters — a deliberate choice to anchor a national program in the country's largest municipal teachers' local. More regional hubs are planned through 2030, turning what could have been a webinar series into a standing network of places teachers physically go. We walk through what actually happens inside that first building in the NYC hub training America's teachers.

Do the arithmetic on the scale and the ambition sharpens. 400,000 teachers over five years is 80,000 a year — a genuinely large annual throughput for in-person and hybrid professional development. Hitting it means standardization: shared materials, a repeatable curriculum, a train-the-trainer multiplier. That standardization is the program's power and its risk at once, a tension we take up in 400,000 teachers, one AI curriculum.

What the money actually buys each side

Strip it to incentives and the deal is legible.

For the union, $23 million funds a members' benefit it could never self-finance at this scale, and — more importantly to its leadership — it puts educators on the designing side of AI training rather than the receiving side. The AFT's repeated framing is "teachers in the driver's seat": members help shape curriculum, test tools, and draft guardrails. Whether that framing survives contact with three vendors' interests is the subject of why a union took AI vendor money.

For the companies, $23 million split three ways is remarkably cheap for what it secures: goodwill with organized labor, a friendly channel into K-12, and — through in-kind compute and engineering — a hand in what "AI literacy" comes to mean for a tenth of the nation's teachers. Coverage in Time situated the deal inside a broader race to define AI's role in schools before the norms harden. That's the real prize. The first mover to shape how teachers think about classroom AI shapes the market for a decade.

The honest read is that this is neither pure philanthropy nor pure land-grab. It's a coalition where the union gets scale and voice it couldn't buy, and the vendors get influence and access they couldn't otherwise earn cheaply — a trade each side entered with open eyes. The $23 million isn't the story. What the two halves of OpenAI's $10 million reveal about the difference between funding a program and shaping it — that's the story.

---

Part 57 of 100 in the ChatGPT for Teachers series. Previously: FERPA and the "School Official" Clause, Decoded. Next: 400,000 Teachers, One AI Curriculum. Browse more builder insights or explore AI skills for education at aiskill.market.